Complexity Explorer Santa Few Institute

Lecture: Artificial Intelligence

Lead instructor:

Your progress is not being saved! Enroll now or log in to track your progress or submit homework.

4.1 Helping AI become "intelligent" » Questions

AI researchers  continue to pursue general intelligence by providing machines with varied ways of interacting with the real (or similated-real) world so that they can "learn by doing" - and learn from experts - just like humans do. It remains challenging to determine exactly what AIs are learning when they succeed at a task. For instance, how much of what AlphaStar learns playing StarCraft can it apply to other games? And measures of AI "intelligence" may still reflect a bias of the human user/programmer.

  • How would you measure the "intelligence" of an AI?
  • Does how we measure AI "intelligence" indicate any problems with the ways we measure human intelligence?
  • What is a simple, real-world task that AlphaStar, the StarCraft master AI, should be able to accomplish if it has learned generalizable knowledge from the game?
  • Is it fair that AlphaStar "stole" its game play strategy from human players?
  • If you were a passionate chess, Go, or StarCraft player and AIs continue to be among the best players in the world, would this lessen your interest in playing these games? Would you want to play against an AI? Would you hope to beat an AI?
  • Why should AIs become "intelligent"? Why aren't narrow AIs sufficient?
  • What other games would you propose for an AI in order to help it gain knowledge of the real world?
  • What might be a good way for an AI to learn human psychology and behavior?
  • Is ethical behavior and decision-making an integral part of being "intelligent"?